sketch-preproc 0.2.1

Sketch preprocessing pipelines with edge detection presets

Sketch Preprocessing Package

A Python package providing sketch preprocessing pipelines for edge detection and mask extraction with a spectrum of edge detection presets.

⚠️ Note: The V4 pipeline is deprecated and will be removed in v1.0.0. Please use the V3 pipeline (default) for all new projects.

Installation

# For development
pip install -e ./sketch_preproc

# For production
pip install sketch-preproc

Quick Start

from sketch_preproc import preprocess

# Use preset 5 (balanced, general purpose) - V3 is the default
result = preprocess(ref_image, user_image, preset=5)

# Access the masks
primary_mask = result["primary_mask"]
detail_mask = result["detail_mask"]
shade_mask = result["shade_mask"]  # Always None for V3

Edge Detection Spectrum Presets

The package includes 10 presets ranging from minimal to maximal edge detection:

1. Bare bones - Absolute minimal, only strongest edges
2. Minimal - Very fine lines, architectural/technical
3. Fine - Clean sketches with fine lines
4. Balanced fine - Good for simple subjects
5. Balanced - General purpose default
6. Balanced detailed - Good for complex subjects
7. Detailed - Captures interior details
8. Very detailed - Includes subtle features
9. Comprehensive - Captures almost everything
10. Maximal - Maximum detail, may include noise

Usage Examples

Using Numeric Presets

from sketch_preproc import preprocess
import cv2

# Load images
ref_img = cv2.imread("reference.jpg")
user_img = cv2.imread("sketch.jpg")

# Minimal edges (preset 1)
result_minimal = preprocess(ref_img, user_img, preset=1)

# Balanced for general use (preset 5)
result_balanced = preprocess(ref_img, user_img, preset=5)

# Maximum detail (preset 10)
result_maximal = preprocess(ref_img, user_img, preset=10)

# Save results
cv2.imwrite("edges_minimal.png", result_minimal["primary_mask"])
cv2.imwrite("edges_balanced.png", result_balanced["primary_mask"])
cv2.imwrite("edges_maximal.png", result_maximal["primary_mask"])

Choosing Pipeline Version

# Use V3 pipeline (default, recommended)
result = preprocess(ref_img, user_img, preset=5)

# Use V4 pipeline (deprecated, not recommended)
import warnings
warnings.filterwarnings("ignore", category=DeprecationWarning)  # Suppress warning
result_v4 = preprocess(ref_img, user_img, pipeline="v4", preset=7)

Note: V4 pipeline is deprecated. It provides an additional shade mask but at the cost of slower processing and potentially noisier results. V3 is recommended for all use cases.

Using Named Presets

# Use specific named preset
result = preprocess(ref_img, user_img, preset="spectrum_03_fine")

# Use default presets
result_v3 = preprocess(ref_img, user_img, preset="default_v3")
result_v4 = preprocess(ref_img, user_img, preset="default_v4")

Custom Configuration

from sketch_preproc.common.config import PreprocCfg

# Create custom config based on preset 5
config = PreprocCfg(
    alpha=0.7,
    tau_primary=0.30,
    tau_detail=0.20,
    tau_lo_primary=0.18,
    tau_lo_detail=0.12
)

result = preprocess(ref_img, user_img, config=config)

List Available Presets

from sketch_preproc import list_presets

# List all presets
all_presets = list_presets()
for key, desc in all_presets.items():
    print(f"{key}: {desc}")

# List only V3 presets (recommended)
v3_presets = list_presets(pipeline="v3")

Choosing the Right Preset

• Presets 1-3: Use for technical drawings, architectural sketches, or when you need very clean lines
• Presets 4-6: General purpose, good balance between detail and noise
• Presets 7-9: Use for artistic sketches with lots of detail or texture
• Preset 10: Maximum detail extraction, useful for analysis but may include noise

Pipeline Information

V3 Pipeline (Recommended)

• Produces two masks: primary and detail
• Faster processing
• Cleaner results with less noise
• Suitable for all use cases
• This is the default and recommended pipeline

V4 Pipeline (Deprecated)

• ⚠️ Deprecated: Will be removed in version 1.0.0
• Produces three masks: primary, detail, and shade
• Slower processing due to additional preprocessing steps
• May produce noisier results
• Only use if you specifically need the shade mask for legacy compatibility

Input Formats

The package accepts multiple input formats:

• File paths (string or Path object)
• NumPy arrays (BGR format)
• Raw bytes (JPEG/PNG encoded)

# From file
result = preprocess("ref.jpg", "sketch.png", preset=5)

# From numpy array
import cv2
ref_img = cv2.imread("ref.jpg")
user_img = cv2.imread("sketch.jpg")
result = preprocess(ref_img, user_img, preset=5)

# From bytes
with open("ref.jpg", "rb") as f:
    ref_bytes = f.read()
with open("sketch.jpg", "rb") as f:
    user_bytes = f.read()
result = preprocess(ref_bytes, user_bytes, preset=5)

Output Format

All pipelines return a dictionary with:

• primary_mask: Primary edge mask (uint8 numpy array)
• detail_mask: Detail edge mask (uint8 numpy array)
• shade_mask: Shade mask (uint8 numpy array, None for V3)
• debug: Dictionary with intermediate processing results

Requirements

• Python 3.8+
• OpenCV with contrib modules (opencv-contrib-python)
• PyTorch
• See pyproject.toml for full dependencies